Assignment Day 8

Assignment

Short objective

The objective of this report is to analyze the incidence rates of Guillain-Barré Syndrom (gbs) and anaphylactic shock using the PHARMO database, and to present the results in a clear and concise manner.

Methods

PHARMO

The data for this analysis was obtained from the PHARMO database, which is a research institute in the Netherlands that maintains a database of electronic medical records. The Pharmo Data Network is a population-based network of healthcare databases and combines data from different healthcare settings in the Netherlands. These different data sources are linked on a patient level through validated algorithms.

Study population

For this research particularly, individuals who were in the study between 01/01/2017 and 31/12/2020 are included. In addition, the study’s inclusion and exclusion criteria consist of three requirements. The first requirement is that individuals must have at least one year of follow-up before the study’s start date. The second requirement is that individuals must have at least one day of follow-up after the study’s start date. Lastly, the study will not consider repeated events, and individuals will be censored at the first occurrence of an event. After fulfilling the aforementioned requirements, we have a study population of N = 675

Data Preperation

Prior to the analysis, various tables are prepared and then combined in order to obtain the incidence rates. First the OBSERVATION_PERIOD table was loaded and individuals who are not in the above mentioned study population, will be removed.

After doing this, the PERSONS table was loaded. In this table, an age variable is made with the birth data variable. The age was created with the start of the study period as reference. with this age variable, the following age bands are created: 0-19, 20-39, 40-59, 60-79. The age band 80+ is not created, because the there are no individuals who are older than 67.

The next step in the date preparation, was merging the OBSERVATION_PERIODS and PERSON table. We only kept the individuals who are in the OBSERVATION_PERIODS table.

When the aforementioned tables are merged, Concept sets were created. These sets were created from diagnostic codes from the codesheet. All diagnostic codes and vocabulary of the events used to make incidence rates can be found in Table 1

When the aforementioned tables are merged, Concept sets were created. These sets were created from diagnostic codes from the codesheet. All diagnostic codes and vocabulary of the events we will make incidence rates of can be found in Table 1.

Figure 1: ANAPHYL and GBS event codes

Lastly, the EVENTS table is loaded, and merged with the event codes. If the persons had an event before the start of the study, they are removed from the dataset. If a person has more than one event, the second event is removed.

Analyis

For the analysis, incidence rates per 1000 person years for the study population are created per event (i.e. Anaphylactic Shock Narrow, Anaphylactic shock broad, and gbs). The formula for the incidence rates can be found below:

\[ \text{Incidence Rates overall} = \frac{Total \ number \ of \ events}{Total \ person-years} * 1000 \]

Afterwards, the incidence rates are stratified by year and age-bands. The formulas for these can be found below: \[ \text{Incidence rates by year} = \frac{total \ number \ of \ events \ for \ each \ year} { person-years \ for \ each \ year} * 1000 \]

\[ \text{Incidence rates by year by age bands} = \frac{Total \ number \ of \ events for \ each \ year \ per \ age \ band}{person-years \ for \ each \ year \ per \ age \ band} * 1000 \]

To assess the technical validation, confidence intervals are used. Exact Poisson confidence limits for the estimated rate are found as the Poisson means, for distributions with the observed number of events and probabilities relevant to the chosen confidence level, divided by time at risk. The relationship between the Poisson and chi-square distributions is employed here (Ulm, 1990):

\[Y_l = \frac{\chi^2_{2Y , a/2}}{2}\] \[Y_u=\frac{\chi^2_2(Y+1), 1-a/2}{2}\]

where Y is the observed number of events, \(Y_l\) and \(Y_u\) are lower and upper confidence limits for Y respectively, \(\chi^2_{v,a}\) is the chi-square quantile for upper tail probability on ν degrees of freedom.

Results

The forest plot figures (see Figure 2), show the incidence rates. A total of 86 events occurred. Of these events 28 were anaphylaxis narrow, 22 were anaphylaxis possible and 36 were GBS. When stratified in to age bands, the person years become lower and there is low confidence. Especially in the high and low age bands. Most events occurred in the first three years. In these years each type of event status had around 15 to 20 occurred events. The last year of the study has almost no events and bad incidence rate confidence interval.

Forest Plot Figures

Total

Figure 2: Forest plots of total IR

2017

Figure 3: forest plots data from 2017

2018

Figure 4: Forest Plot data from 2018

2019

Figure 5: Forest Plot data from 2019

2020

Figure 6: Forest plot data from 2020

Lessons learned

We have learned to take data from various tables, and structure it to a dataset which can be analysed. In addition, we have learned that a lot of data is needed to make inference about incidence rates. That is, when data needs to be stratified. We discovered that the distribution of the data becomes skewed when stratified in different age bands. In this case the incidence over a 1000 person years confidence interval will get wider and thus make it harder to indicate the true incidence rate. More data from different sources could help in creating a better distributed stratification. This can be accomplished by using data from a Common Data Model. The data could also have some bias. All of the data is collected using health registries and (routine) care data. This means that every person in the dataset, has had contact with an health institution. Thus, perfectly healthy people with no events will not be included in the dataset.

Annexes

Annex 1 Results Table

Annex 2 Day 6